New Jersey bill would make chatbot makers liable for harm to users

A New Jersey bill would make companies that offer AI chatbots legally responsible for any injury a user suffers from the chatbot, even if the company took every reasonable precaution. Bill A5607 was reported favorably by an Assembly committee on Oct. 19 and now goes to the Assembly Appropriations Committee. It is not law, and none of its rules are in effect.

What the bill would do

The bill would treat a chatbot as a "product." That means people who say they were hurt by one could bring the same kind of claim they would bring over a faulty appliance. Providers would also have a "duty to ensure that use of their chatbot does not cause injury to a user."

A provider would be liable even if it used all reasonable care in designing and distributing the chatbot. It would also be liable if it never gave the chatbot to the user directly or had no contract with that user. Lawyers call this strict liability, which means fault does not have to be proven.

Who it could reach

The bill defines a chatbot broadly: any software that simulates human conversation through text, images, video or audio. A provider is anyone who creates, distributes or otherwise makes available a chatbot. That could cover not only AI developers but also businesses that put a chatbot on their website or app, such as a customer service assistant.

Other states draw the line differently. California's SB 243 and New York's AI Companion Models Law apply to companion chatbots, and California expressly excludes bots used only for customer service. A5607, as introduced, has no similar carve-out. Committee discussion also noted that the bill's geographic reach is unclear.

Privacy duties

Chatbot providers would be treated as "controllers" under the New Jersey Data Privacy Act, which adds that law's obligations to the bill's own. The bill also sets a high bar for user consent:

• A clear, standalone disclosure in easy-to-understand language, offered in each language the chatbot uses
• A refuse option at least as prominent as the consent option, and no more steps to reach
• No consent inferred from inaction or continued use, or from accepting general terms of use

Providers would also have to post information about their chatbot on their website and let users access and download their chat logs. A written data security program would have to be publicly posted as well.

What is still unclear

Current New Jersey product law requires a person to prove a defect, and it offers sellers certain defenses. The bill does not say whether proof of a defect would be required or whether those defenses would survive. Supporters say it is based on a model bill from EPIC, the Consumer Federation of America and Fairplay.

Courts have only begun to weigh in. A federal court in Florida let claims against a companion chatbot maker proceed in May 2025, and that case was later settled. Similar claims against OpenAI are pending in San Francisco. First Amendment challenges to the bill would remain possible, and its text may change.